Dongyan Zhao (赵东岩)

Peking University professor and Engram coauthor. PKU AI and ICST pages identify him as professor at Wangxuan Institute of Computer Technology with NLP and semantic data management interests; OpenReview and Scholar corroborate the PKU identity. No high-confidence personal X account was found.

2 archived notes X: not-found Homepage

Jun 28, 2026

DSpark: Confidence Scheduled Speculative Decoding with Semi Autoregressive Generation

DSpark 是 DeepSeek 把并行 drafter 推向生产 serving 的一套完整方案：用 DFlash 式 parallel backbone 先一次生成长候选块，再用低秩 Markov head 注入块内局部自回归依赖，随后用 calibrated confidence head 和硬件感知 prefix scheduler 按请求与负载动态裁剪 target verification 长度。离线 Qwen3 / G...

2026-06-27-dspark-confidence-scheduled-speculative-decoding Speculative Decoding Multi-Token Prediction Serving Runtime

Jun 24, 2026

Conditional Memory via Scalable Lookup: A New Axis of Sparsity for Large Language Models

Engram 把大模型稀疏性从 MoE (Mixture of Experts，混合专家，用条件计算扩大 total 参数并控制 active compute) 扩展到 conditional memory：用 hashed N gram lookup 存静态局部模式，用 context aware gate 决定是否把查到的记忆注入 hidden state；在 iso parameter / iso FLOPs 的 27B MoE...

2601.07372-conditional-memory-engram-scalable-lookup Memory Architecture MoE Architecture

Related Notes

DSpark: Confidence Scheduled Speculative Decoding with Semi Autoregressive Generation

Conditional Memory via Scalable Lookup: A New Axis of Sparsity for Large Language Models